fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/fs.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/nfs_idmap.h>
  46 #include <linux/kthread.h>
  47 #include <linux/module.h>
  48 #include <linux/random.h>
  49 #include <linux/ratelimit.h>
  50 #include <linux/workqueue.h>
  51 #include <linux/bitops.h>
  52
  53 #include "nfs4_fs.h"
  54 #include "callback.h"
  55 #include "delegation.h"
  56 #include "internal.h"
  57 #include "pnfs.h"
  58
  59 #define OPENOWNER_POOL_SIZE     8
  60
  61 const nfs4_stateid zero_stateid;
  62
  63 static LIST_HEAD(nfs4_clientid_list);
  64
  65 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  66 {
  67         struct nfs4_setclientid_res clid;
  68         unsigned short port;
  69         int status;
  70
  71         port = nfs_callback_tcpport;
  72         if (clp->cl_addr.ss_family == AF_INET6)
  73                 port = nfs_callback_tcpport6;
  74
  75         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  76         if (status != 0)
  77                 goto out;
  78         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  79         if (status != 0)
  80                 goto out;
  81         clp->cl_clientid = clid.clientid;
  82         nfs4_schedule_state_renewal(clp);
  83 out:
  84         return status;
  85 }
  86
  87 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
  88 {
  89         struct rpc_cred *cred = NULL;
  90
  91         if (clp->cl_machine_cred != NULL)
  92                 cred = get_rpccred(clp->cl_machine_cred);
  93         return cred;
  94 }
  95
  96 static void nfs4_clear_machine_cred(struct nfs_client *clp)
  97 {
  98         struct rpc_cred *cred;
  99
 100         spin_lock(&clp->cl_lock);
 101         cred = clp->cl_machine_cred;
 102         clp->cl_machine_cred = NULL;
 103         spin_unlock(&clp->cl_lock);
 104         if (cred != NULL)
 105                 put_rpccred(cred);
 106 }
 107
 108 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 109 {
 110         struct nfs4_state_owner *sp;
 111         struct rb_node *pos;
 112         struct rpc_cred *cred = NULL;
 113
 114         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
 115                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 116                 if (list_empty(&sp->so_states))
 117                         continue;
 118                 cred = get_rpccred(sp->so_cred);
 119                 break;
 120         }
 121         return cred;
 122 }
 123
 124 #if defined(CONFIG_NFS_V4_1)
 125
 126 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 127 {
 128         int status;
 129         struct nfs_fsinfo fsinfo;
 130
 131         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 132         if (status == 0) {
 133                 /* Update lease time and schedule renewal */
 134                 spin_lock(&clp->cl_lock);
 135                 clp->cl_lease_time = fsinfo.lease_time * HZ;
 136                 clp->cl_last_renewal = jiffies;
 137                 spin_unlock(&clp->cl_lock);
 138
 139                 nfs4_schedule_state_renewal(clp);
 140         }
 141
 142         return status;
 143 }
 144
 145 static void nfs4_end_drain_session(struct nfs_client *clp)
 146 {
 147         struct nfs4_session *ses = clp->cl_session;
 148         int max_slots;
 149
 150         if (ses == NULL)
 151                 return;
 152         if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 153                 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
 154                 max_slots = ses->fc_slot_table.max_slots;
 155                 while (max_slots--) {
 156                         struct rpc_task *task;
 157
 158                         task = rpc_wake_up_next(&ses->fc_slot_table.
 159                                                 slot_tbl_waitq);
 160                         if (!task)
 161                                 break;
 162                         rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
 163                 }
 164                 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
 165         }
 166 }
 167
 168 static int nfs4_begin_drain_session(struct nfs_client *clp)
 169 {
 170         struct nfs4_session *ses = clp->cl_session;
 171         struct nfs4_slot_table *tbl = &ses->fc_slot_table;
 172
 173         spin_lock(&tbl->slot_tbl_lock);
 174         set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
 175         if (tbl->highest_used_slotid != -1) {
 176                 INIT_COMPLETION(ses->complete);
 177                 spin_unlock(&tbl->slot_tbl_lock);
 178                 return wait_for_completion_interruptible(&ses->complete);
 179         }
 180         spin_unlock(&tbl->slot_tbl_lock);
 181         return 0;
 182 }
 183
 184 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 185 {
 186         int status;
 187
 188         nfs4_begin_drain_session(clp);
 189         status = nfs4_proc_exchange_id(clp, cred);
 190         if (status != 0)
 191                 goto out;
 192         status = nfs4_proc_create_session(clp);
 193         if (status != 0)
 194                 goto out;
 195         status = nfs4_set_callback_sessionid(clp);
 196         if (status != 0) {
 197                 printk(KERN_WARNING "Sessionid not set. No callback service\n");
 198                 nfs_callback_down(1);
 199                 status = 0;
 200         }
 201         nfs41_setup_state_renewal(clp);
 202         nfs_mark_client_ready(clp, NFS_CS_READY);
 203 out:
 204         return status;
 205 }
 206
 207 struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
 208 {
 209         struct rpc_cred *cred;
 210
 211         spin_lock(&clp->cl_lock);
 212         cred = nfs4_get_machine_cred_locked(clp);
 213         spin_unlock(&clp->cl_lock);
 214         return cred;
 215 }
 216
 217 #endif /* CONFIG_NFS_V4_1 */
 218
 219 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
 220 {
 221         struct nfs4_state_owner *sp;
 222         struct rb_node *pos;
 223         struct rpc_cred *cred;
 224
 225         spin_lock(&clp->cl_lock);
 226         cred = nfs4_get_machine_cred_locked(clp);
 227         if (cred != NULL)
 228                 goto out;
 229         pos = rb_first(&clp->cl_state_owners);
 230         if (pos != NULL) {
 231                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
 232                 cred = get_rpccred(sp->so_cred);
 233         }
 234 out:
 235         spin_unlock(&clp->cl_lock);
 236         return cred;
 237 }
 238
 239 static void nfs_alloc_unique_id(struct rb_root *root, struct nfs_unique_id *new,
 240                 __u64 minval, int maxbits)
 241 {
 242         struct rb_node **p, *parent;
 243         struct nfs_unique_id *pos;
 244         __u64 mask = ~0ULL;
 245
 246         if (maxbits < 64)
 247                 mask = (1ULL << maxbits) - 1ULL;
 248
 249         /* Ensure distribution is more or less flat */
 250         get_random_bytes(&new->id, sizeof(new->id));
 251         new->id &= mask;
 252         if (new->id < minval)
 253                 new->id += minval;
 254 retry:
 255         p = &root->rb_node;
 256         parent = NULL;
 257
 258         while (*p != NULL) {
 259                 parent = *p;
 260                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 261
 262                 if (new->id < pos->id)
 263                         p = &(*p)->rb_left;
 264                 else if (new->id > pos->id)
 265                         p = &(*p)->rb_right;
 266                 else
 267                         goto id_exists;
 268         }
 269         rb_link_node(&new->rb_node, parent, p);
 270         rb_insert_color(&new->rb_node, root);
 271         return;
 272 id_exists:
 273         for (;;) {
 274                 new->id++;
 275                 if (new->id < minval || (new->id & mask) != new->id) {
 276                         new->id = minval;
 277                         break;
 278                 }
 279                 parent = rb_next(parent);
 280                 if (parent == NULL)
 281                         break;
 282                 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
 283                 if (new->id < pos->id)
 284                         break;
 285         }
 286         goto retry;
 287 }
 288
 289 static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
 290 {
 291         rb_erase(&id->rb_node, root);
 292 }
 293
 294 static struct nfs4_state_owner *
 295 nfs4_find_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 296 {
 297         struct nfs_client *clp = server->nfs_client;
 298         struct rb_node **p = &clp->cl_state_owners.rb_node,
 299                        *parent = NULL;
 300         struct nfs4_state_owner *sp, *res = NULL;
 301
 302         while (*p != NULL) {
 303                 parent = *p;
 304                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 305
 306                 if (server < sp->so_server) {
 307                         p = &parent->rb_left;
 308                         continue;
 309                 }
 310                 if (server > sp->so_server) {
 311                         p = &parent->rb_right;
 312                         continue;
 313                 }
 314                 if (cred < sp->so_cred)
 315                         p = &parent->rb_left;
 316                 else if (cred > sp->so_cred)
 317                         p = &parent->rb_right;
 318                 else {
 319                         atomic_inc(&sp->so_count);
 320                         res = sp;
 321                         break;
 322                 }
 323         }
 324         return res;
 325 }
 326
 327 static struct nfs4_state_owner *
 328 nfs4_insert_state_owner(struct nfs_client *clp, struct nfs4_state_owner *new)
 329 {
 330         struct rb_node **p = &clp->cl_state_owners.rb_node,
 331                        *parent = NULL;
 332         struct nfs4_state_owner *sp;
 333
 334         while (*p != NULL) {
 335                 parent = *p;
 336                 sp = rb_entry(parent, struct nfs4_state_owner, so_client_node);
 337
 338                 if (new->so_server < sp->so_server) {
 339                         p = &parent->rb_left;
 340                         continue;
 341                 }
 342                 if (new->so_server > sp->so_server) {
 343                         p = &parent->rb_right;
 344                         continue;
 345                 }
 346                 if (new->so_cred < sp->so_cred)
 347                         p = &parent->rb_left;
 348                 else if (new->so_cred > sp->so_cred)
 349                         p = &parent->rb_right;
 350                 else {
 351                         atomic_inc(&sp->so_count);
 352                         return sp;
 353                 }
 354         }
 355         nfs_alloc_unique_id(&clp->cl_openowner_id, &new->so_owner_id, 1, 64);
 356         rb_link_node(&new->so_client_node, parent, p);
 357         rb_insert_color(&new->so_client_node, &clp->cl_state_owners);
 358         return new;
 359 }
 360
 361 static void
 362 nfs4_remove_state_owner(struct nfs_client *clp, struct nfs4_state_owner *sp)
 363 {
 364         if (!RB_EMPTY_NODE(&sp->so_client_node))
 365                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 366         nfs_free_unique_id(&clp->cl_openowner_id, &sp->so_owner_id);
 367 }
 368
 369 /*
 370  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 371  * create a new state_owner.
 372  *
 373  */
 374 static struct nfs4_state_owner *
 375 nfs4_alloc_state_owner(void)
 376 {
 377         struct nfs4_state_owner *sp;
 378
 379         sp = kzalloc(sizeof(*sp),GFP_NOFS);
 380         if (!sp)
 381                 return NULL;
 382         spin_lock_init(&sp->so_lock);
 383         INIT_LIST_HEAD(&sp->so_states);
 384         rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
 385         sp->so_seqid.sequence = &sp->so_sequence;
 386         spin_lock_init(&sp->so_sequence.lock);
 387         INIT_LIST_HEAD(&sp->so_sequence.list);
 388         atomic_set(&sp->so_count, 1);
 389         return sp;
 390 }
 391
 392 static void
 393 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 394 {
 395         if (!RB_EMPTY_NODE(&sp->so_client_node)) {
 396                 struct nfs_client *clp = sp->so_server->nfs_client;
 397
 398                 spin_lock(&clp->cl_lock);
 399                 rb_erase(&sp->so_client_node, &clp->cl_state_owners);
 400                 RB_CLEAR_NODE(&sp->so_client_node);
 401                 spin_unlock(&clp->cl_lock);
 402         }
 403 }
 404
 405 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
 406 {
 407         struct nfs_client *clp = server->nfs_client;
 408         struct nfs4_state_owner *sp, *new;
 409
 410         spin_lock(&clp->cl_lock);
 411         sp = nfs4_find_state_owner(server, cred);
 412         spin_unlock(&clp->cl_lock);
 413         if (sp != NULL)
 414                 return sp;
 415         new = nfs4_alloc_state_owner();
 416         if (new == NULL)
 417                 return NULL;
 418         new->so_server = server;
 419         new->so_cred = cred;
 420         spin_lock(&clp->cl_lock);
 421         sp = nfs4_insert_state_owner(clp, new);
 422         spin_unlock(&clp->cl_lock);
 423         if (sp == new)
 424                 get_rpccred(cred);
 425         else {
 426                 rpc_destroy_wait_queue(&new->so_sequence.wait);
 427                 kfree(new);
 428         }
 429         return sp;
 430 }
 431
 432 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 433 {
 434         struct nfs_client *clp = sp->so_server->nfs_client;
 435         struct rpc_cred *cred = sp->so_cred;
 436
 437         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 438                 return;
 439         nfs4_remove_state_owner(clp, sp);
 440         spin_unlock(&clp->cl_lock);
 441         rpc_destroy_wait_queue(&sp->so_sequence.wait);
 442         put_rpccred(cred);
 443         kfree(sp);
 444 }
 445
 446 static struct nfs4_state *
 447 nfs4_alloc_open_state(void)
 448 {
 449         struct nfs4_state *state;
 450
 451         state = kzalloc(sizeof(*state), GFP_NOFS);
 452         if (!state)
 453                 return NULL;
 454         atomic_set(&state->count, 1);
 455         INIT_LIST_HEAD(&state->lock_states);
 456         spin_lock_init(&state->state_lock);
 457         seqlock_init(&state->seqlock);
 458         return state;
 459 }
 460
 461 void
 462 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 463 {
 464         if (state->state == fmode)
 465                 return;
 466         /* NB! List reordering - see the reclaim code for why.  */
 467         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 468                 if (fmode & FMODE_WRITE)
 469                         list_move(&state->open_states, &state->owner->so_states);
 470                 else
 471                         list_move_tail(&state->open_states, &state->owner->so_states);
 472         }
 473         state->state = fmode;
 474 }
 475
 476 static struct nfs4_state *
 477 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 478 {
 479         struct nfs_inode *nfsi = NFS_I(inode);
 480         struct nfs4_state *state;
 481
 482         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 483                 if (state->owner != owner)
 484                         continue;
 485                 if (atomic_inc_not_zero(&state->count))
 486                         return state;
 487         }
 488         return NULL;
 489 }
 490
 491 static void
 492 nfs4_free_open_state(struct nfs4_state *state)
 493 {
 494         kfree(state);
 495 }
 496
 497 struct nfs4_state *
 498 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 499 {
 500         struct nfs4_state *state, *new;
 501         struct nfs_inode *nfsi = NFS_I(inode);
 502
 503         spin_lock(&inode->i_lock);
 504         state = __nfs4_find_state_byowner(inode, owner);
 505         spin_unlock(&inode->i_lock);
 506         if (state)
 507                 goto out;
 508         new = nfs4_alloc_open_state();
 509         spin_lock(&owner->so_lock);
 510         spin_lock(&inode->i_lock);
 511         state = __nfs4_find_state_byowner(inode, owner);
 512         if (state == NULL && new != NULL) {
 513                 state = new;
 514                 state->owner = owner;
 515                 atomic_inc(&owner->so_count);
 516                 list_add(&state->inode_states, &nfsi->open_states);
 517                 state->inode = igrab(inode);
 518                 spin_unlock(&inode->i_lock);
 519                 /* Note: The reclaim code dictates that we add stateless
 520                  * and read-only stateids to the end of the list */
 521                 list_add_tail(&state->open_states, &owner->so_states);
 522                 spin_unlock(&owner->so_lock);
 523         } else {
 524                 spin_unlock(&inode->i_lock);
 525                 spin_unlock(&owner->so_lock);
 526                 if (new)
 527                         nfs4_free_open_state(new);
 528         }
 529 out:
 530         return state;
 531 }
 532
 533 void nfs4_put_open_state(struct nfs4_state *state)
 534 {
 535         struct inode *inode = state->inode;
 536         struct nfs4_state_owner *owner = state->owner;
 537
 538         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 539                 return;
 540         spin_lock(&inode->i_lock);
 541         list_del(&state->inode_states);
 542         list_del(&state->open_states);
 543         spin_unlock(&inode->i_lock);
 544         spin_unlock(&owner->so_lock);
 545         iput(inode);
 546         nfs4_free_open_state(state);
 547         nfs4_put_state_owner(owner);
 548 }
 549
 550 /*
 551  * Close the current file.
 552  */
 553 static void __nfs4_close(struct path *path, struct nfs4_state *state,
 554                 fmode_t fmode, gfp_t gfp_mask, int wait)
 555 {
 556         struct nfs4_state_owner *owner = state->owner;
 557         int call_close = 0;
 558         fmode_t newstate;
 559
 560         atomic_inc(&owner->so_count);
 561         /* Protect against nfs4_find_state() */
 562         spin_lock(&owner->so_lock);
 563         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 564                 case FMODE_READ:
 565                         state->n_rdonly--;
 566                         break;
 567                 case FMODE_WRITE:
 568                         state->n_wronly--;
 569                         break;
 570                 case FMODE_READ|FMODE_WRITE:
 571                         state->n_rdwr--;
 572         }
 573         newstate = FMODE_READ|FMODE_WRITE;
 574         if (state->n_rdwr == 0) {
 575                 if (state->n_rdonly == 0) {
 576                         newstate &= ~FMODE_READ;
 577                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 578                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 579                 }
 580                 if (state->n_wronly == 0) {
 581                         newstate &= ~FMODE_WRITE;
 582                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 583                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 584                 }
 585                 if (newstate == 0)
 586                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 587         }
 588         nfs4_state_set_mode_locked(state, newstate);
 589         spin_unlock(&owner->so_lock);
 590
 591         if (!call_close) {
 592                 nfs4_put_open_state(state);
 593                 nfs4_put_state_owner(owner);
 594         } else
 595                 nfs4_do_close(path, state, gfp_mask, wait);
 596 }
 597
 598 void nfs4_close_state(struct path *path, struct nfs4_state *state, fmode_t fmode)
 599 {
 600         __nfs4_close(path, state, fmode, GFP_NOFS, 0);
 601 }
 602
 603 void nfs4_close_sync(struct path *path, struct nfs4_state *state, fmode_t fmode)
 604 {
 605         __nfs4_close(path, state, fmode, GFP_KERNEL, 1);
 606 }
 607
 608 /*
 609  * Search the state->lock_states for an existing lock_owner
 610  * that is compatible with current->files
 611  */
 612 static struct nfs4_lock_state *
 613 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 614 {
 615         struct nfs4_lock_state *pos;
 616         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 617                 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
 618                         continue;
 619                 switch (pos->ls_owner.lo_type) {
 620                 case NFS4_POSIX_LOCK_TYPE:
 621                         if (pos->ls_owner.lo_u.posix_owner != fl_owner)
 622                                 continue;
 623                         break;
 624                 case NFS4_FLOCK_LOCK_TYPE:
 625                         if (pos->ls_owner.lo_u.flock_owner != fl_pid)
 626                                 continue;
 627                 }
 628                 atomic_inc(&pos->ls_count);
 629                 return pos;
 630         }
 631         return NULL;
 632 }
 633
 634 /*
 635  * Return a compatible lock_state. If no initialized lock_state structure
 636  * exists, return an uninitialized one.
 637  *
 638  */
 639 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
 640 {
 641         struct nfs4_lock_state *lsp;
 642         struct nfs_client *clp = state->owner->so_server->nfs_client;
 643
 644         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 645         if (lsp == NULL)
 646                 return NULL;
 647         rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
 648         spin_lock_init(&lsp->ls_sequence.lock);
 649         INIT_LIST_HEAD(&lsp->ls_sequence.list);
 650         lsp->ls_seqid.sequence = &lsp->ls_sequence;
 651         atomic_set(&lsp->ls_count, 1);
 652         lsp->ls_state = state;
 653         lsp->ls_owner.lo_type = type;
 654         switch (lsp->ls_owner.lo_type) {
 655         case NFS4_FLOCK_LOCK_TYPE:
 656                 lsp->ls_owner.lo_u.flock_owner = fl_pid;
 657                 break;
 658         case NFS4_POSIX_LOCK_TYPE:
 659                 lsp->ls_owner.lo_u.posix_owner = fl_owner;
 660                 break;
 661         default:
 662                 kfree(lsp);
 663                 return NULL;
 664         }
 665         spin_lock(&clp->cl_lock);
 666         nfs_alloc_unique_id(&clp->cl_lockowner_id, &lsp->ls_id, 1, 64);
 667         spin_unlock(&clp->cl_lock);
 668         INIT_LIST_HEAD(&lsp->ls_locks);
 669         return lsp;
 670 }
 671
 672 static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
 673 {
 674         struct nfs_client *clp = lsp->ls_state->owner->so_server->nfs_client;
 675
 676         spin_lock(&clp->cl_lock);
 677         nfs_free_unique_id(&clp->cl_lockowner_id, &lsp->ls_id);
 678         spin_unlock(&clp->cl_lock);
 679         rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
 680         kfree(lsp);
 681 }
 682
 683 /*
 684  * Return a compatible lock_state. If no initialized lock_state structure
 685  * exists, return an uninitialized one.
 686  *
 687  */
 688 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
 689 {
 690         struct nfs4_lock_state *lsp, *new = NULL;
 691
 692         for(;;) {
 693                 spin_lock(&state->state_lock);
 694                 lsp = __nfs4_find_lock_state(state, owner, pid, type);
 695                 if (lsp != NULL)
 696                         break;
 697                 if (new != NULL) {
 698                         list_add(&new->ls_locks, &state->lock_states);
 699                         set_bit(LK_STATE_IN_USE, &state->flags);
 700                         lsp = new;
 701                         new = NULL;
 702                         break;
 703                 }
 704                 spin_unlock(&state->state_lock);
 705                 new = nfs4_alloc_lock_state(state, owner, pid, type);
 706                 if (new == NULL)
 707                         return NULL;
 708         }
 709         spin_unlock(&state->state_lock);
 710         if (new != NULL)
 711                 nfs4_free_lock_state(new);
 712         return lsp;
 713 }
 714
 715 /*
 716  * Release reference to lock_state, and free it if we see that
 717  * it is no longer in use
 718  */
 719 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 720 {
 721         struct nfs4_state *state;
 722
 723         if (lsp == NULL)
 724                 return;
 725         state = lsp->ls_state;
 726         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 727                 return;
 728         list_del(&lsp->ls_locks);
 729         if (list_empty(&state->lock_states))
 730                 clear_bit(LK_STATE_IN_USE, &state->flags);
 731         spin_unlock(&state->state_lock);
 732         if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
 733                 nfs4_release_lockowner(lsp);
 734         nfs4_free_lock_state(lsp);
 735 }
 736
 737 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 738 {
 739         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 740
 741         dst->fl_u.nfs4_fl.owner = lsp;
 742         atomic_inc(&lsp->ls_count);
 743 }
 744
 745 static void nfs4_fl_release_lock(struct file_lock *fl)
 746 {
 747         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 748 }
 749
 750 static const struct file_lock_operations nfs4_fl_lock_ops = {
 751         .fl_copy_lock = nfs4_fl_copy_lock,
 752         .fl_release_private = nfs4_fl_release_lock,
 753 };
 754
 755 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 756 {
 757         struct nfs4_lock_state *lsp;
 758
 759         if (fl->fl_ops != NULL)
 760                 return 0;
 761         if (fl->fl_flags & FL_POSIX)
 762                 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
 763         else if (fl->fl_flags & FL_FLOCK)
 764                 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
 765         else
 766                 return -EINVAL;
 767         if (lsp == NULL)
 768                 return -ENOMEM;
 769         fl->fl_u.nfs4_fl.owner = lsp;
 770         fl->fl_ops = &nfs4_fl_lock_ops;
 771         return 0;
 772 }
 773
 774 /*
 775  * Byte-range lock aware utility to initialize the stateid of read/write
 776  * requests.
 777  */
 778 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
 779 {
 780         struct nfs4_lock_state *lsp;
 781         int seq;
 782
 783         do {
 784                 seq = read_seqbegin(&state->seqlock);
 785                 memcpy(dst, &state->stateid, sizeof(*dst));
 786         } while (read_seqretry(&state->seqlock, seq));
 787         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 788                 return;
 789
 790         spin_lock(&state->state_lock);
 791         lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
 792         if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
 793                 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
 794         spin_unlock(&state->state_lock);
 795         nfs4_put_lock_state(lsp);
 796 }
 797
 798 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
 799 {
 800         struct nfs_seqid *new;
 801
 802         new = kmalloc(sizeof(*new), gfp_mask);
 803         if (new != NULL) {
 804                 new->sequence = counter;
 805                 INIT_LIST_HEAD(&new->list);
 806         }
 807         return new;
 808 }
 809
 810 void nfs_release_seqid(struct nfs_seqid *seqid)
 811 {
 812         if (!list_empty(&seqid->list)) {
 813                 struct rpc_sequence *sequence = seqid->sequence->sequence;
 814
 815                 spin_lock(&sequence->lock);
 816                 list_del_init(&seqid->list);
 817                 spin_unlock(&sequence->lock);
 818                 rpc_wake_up(&sequence->wait);
 819         }
 820 }
 821
 822 void nfs_free_seqid(struct nfs_seqid *seqid)
 823 {
 824         nfs_release_seqid(seqid);
 825         kfree(seqid);
 826 }
 827
 828 /*
 829  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
 830  * failed with a seqid incrementing error -
 831  * see comments nfs_fs.h:seqid_mutating_error()
 832  */
 833 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
 834 {
 835         BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
 836         switch (status) {
 837                 case 0:
 838                         break;
 839                 case -NFS4ERR_BAD_SEQID:
 840                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
 841                                 return;
 842                         printk(KERN_WARNING "NFS: v4 server returned a bad"
 843                                         " sequence-id error on an"
 844                                         " unconfirmed sequence %p!\n",
 845                                         seqid->sequence);
 846                 case -NFS4ERR_STALE_CLIENTID:
 847                 case -NFS4ERR_STALE_STATEID:
 848                 case -NFS4ERR_BAD_STATEID:
 849                 case -NFS4ERR_BADXDR:
 850                 case -NFS4ERR_RESOURCE:
 851                 case -NFS4ERR_NOFILEHANDLE:
 852                         /* Non-seqid mutating errors */
 853                         return;
 854         };
 855         /*
 856          * Note: no locking needed as we are guaranteed to be first
 857          * on the sequence list
 858          */
 859         seqid->sequence->counter++;
 860 }
 861
 862 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
 863 {
 864         struct nfs4_state_owner *sp = container_of(seqid->sequence,
 865                                         struct nfs4_state_owner, so_seqid);
 866         struct nfs_server *server = sp->so_server;
 867
 868         if (status == -NFS4ERR_BAD_SEQID)
 869                 nfs4_drop_state_owner(sp);
 870         if (!nfs4_has_session(server->nfs_client))
 871                 nfs_increment_seqid(status, seqid);
 872 }
 873
 874 /*
 875  * Increment the seqid if the LOCK/LOCKU succeeded, or
 876  * failed with a seqid incrementing error -
 877  * see comments nfs_fs.h:seqid_mutating_error()
 878  */
 879 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
 880 {
 881         nfs_increment_seqid(status, seqid);
 882 }
 883
 884 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
 885 {
 886         struct rpc_sequence *sequence = seqid->sequence->sequence;
 887         int status = 0;
 888
 889         spin_lock(&sequence->lock);
 890         if (list_empty(&seqid->list))
 891                 list_add_tail(&seqid->list, &sequence->list);
 892         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
 893                 goto unlock;
 894         rpc_sleep_on(&sequence->wait, task, NULL);
 895         status = -EAGAIN;
 896 unlock:
 897         spin_unlock(&sequence->lock);
 898         return status;
 899 }
 900
 901 static int nfs4_run_state_manager(void *);
 902
 903 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
 904 {
 905         smp_mb__before_clear_bit();
 906         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
 907         smp_mb__after_clear_bit();
 908         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
 909         rpc_wake_up(&clp->cl_rpcwaitq);
 910 }
 911
 912 /*
 913  * Schedule the nfs_client asynchronous state management routine
 914  */
 915 void nfs4_schedule_state_manager(struct nfs_client *clp)
 916 {
 917         struct task_struct *task;
 918
 919         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
 920                 return;
 921         __module_get(THIS_MODULE);
 922         atomic_inc(&clp->cl_count);
 923         task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
 924                                 rpc_peeraddr2str(clp->cl_rpcclient,
 925                                                         RPC_DISPLAY_ADDR));
 926         if (!IS_ERR(task))
 927                 return;
 928         nfs4_clear_state_manager_bit(clp);
 929         nfs_put_client(clp);
 930         module_put(THIS_MODULE);
 931 }
 932
 933 /*
 934  * Schedule a state recovery attempt
 935  */
 936 void nfs4_schedule_state_recovery(struct nfs_client *clp)
 937 {
 938         if (!clp)
 939                 return;
 940         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
 941                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
 942         nfs4_schedule_state_manager(clp);
 943 }
 944
 945 int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
 946 {
 947
 948         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 949         /* Don't recover state that expired before the reboot */
 950         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
 951                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 952                 return 0;
 953         }
 954         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
 955         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
 956         return 1;
 957 }
 958
 959 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
 960 {
 961         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
 962         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
 963         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
 964         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
 965         return 1;
 966 }
 967
 968 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
 969 {
 970         struct inode *inode = state->inode;
 971         struct nfs_inode *nfsi = NFS_I(inode);
 972         struct file_lock *fl;
 973         int status = 0;
 974
 975         if (inode->i_flock == NULL)
 976                 return 0;
 977
 978         /* Guard against delegation returns and new lock/unlock calls */
 979         down_write(&nfsi->rwsem);
 980         /* Protect inode->i_flock using the BKL */
 981         lock_flocks();
 982         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
 983                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
 984                         continue;
 985                 if (nfs_file_open_context(fl->fl_file)->state != state)
 986                         continue;
 987                 unlock_flocks();
 988                 status = ops->recover_lock(state, fl);
 989                 switch (status) {
 990                         case 0:
 991                                 break;
 992                         case -ESTALE:
 993                         case -NFS4ERR_ADMIN_REVOKED:
 994                         case -NFS4ERR_STALE_STATEID:
 995                         case -NFS4ERR_BAD_STATEID:
 996                         case -NFS4ERR_EXPIRED:
 997                         case -NFS4ERR_NO_GRACE:
 998                         case -NFS4ERR_STALE_CLIENTID:
 999                         case -NFS4ERR_BADSESSION:
1000                         case -NFS4ERR_BADSLOT:
1001                         case -NFS4ERR_BAD_HIGH_SLOT:
1002                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1003                                 goto out;
1004                         default:
1005                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1006                                                 __func__, status);
1007                         case -ENOMEM:
1008                         case -NFS4ERR_DENIED:
1009                         case -NFS4ERR_RECLAIM_BAD:
1010                         case -NFS4ERR_RECLAIM_CONFLICT:
1011                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1012                                 status = 0;
1013                 }
1014                 lock_flocks();
1015         }
1016         unlock_flocks();
1017 out:
1018         up_write(&nfsi->rwsem);
1019         return status;
1020 }
1021
1022 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1023 {
1024         struct nfs4_state *state;
1025         struct nfs4_lock_state *lock;
1026         int status = 0;
1027
1028         /* Note: we rely on the sp->so_states list being ordered
1029          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1030          * states first.
1031          * This is needed to ensure that the server won't give us any
1032          * read delegations that we have to return if, say, we are
1033          * recovering after a network partition or a reboot from a
1034          * server that doesn't support a grace period.
1035          */
1036 restart:
1037         spin_lock(&sp->so_lock);
1038         list_for_each_entry(state, &sp->so_states, open_states) {
1039                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1040                         continue;
1041                 if (state->state == 0)
1042                         continue;
1043                 atomic_inc(&state->count);
1044                 spin_unlock(&sp->so_lock);
1045                 status = ops->recover_open(sp, state);
1046                 if (status >= 0) {
1047                         status = nfs4_reclaim_locks(state, ops);
1048                         if (status >= 0) {
1049                                 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1050                                         if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1051                                                 printk("%s: Lock reclaim failed!\n",
1052                                                         __func__);
1053                                 }
1054                                 nfs4_put_open_state(state);
1055                                 goto restart;
1056                         }
1057                 }
1058                 switch (status) {
1059                         default:
1060                                 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1061                                                 __func__, status);
1062                         case -ENOENT:
1063                         case -ENOMEM:
1064                         case -ESTALE:
1065                                 /*
1066                                  * Open state on this file cannot be recovered
1067                                  * All we can do is revert to using the zero stateid.
1068                                  */
1069                                 memset(state->stateid.data, 0,
1070                                         sizeof(state->stateid.data));
1071                                 /* Mark the file as being 'closed' */
1072                                 state->state = 0;
1073                                 break;
1074                         case -EKEYEXPIRED:
1075                                 /*
1076                                  * User RPCSEC_GSS context has expired.
1077                                  * We cannot recover this stateid now, so
1078                                  * skip it and allow recovery thread to
1079                                  * proceed.
1080                                  */
1081                                 break;
1082                         case -NFS4ERR_ADMIN_REVOKED:
1083                         case -NFS4ERR_STALE_STATEID:
1084                         case -NFS4ERR_BAD_STATEID:
1085                         case -NFS4ERR_RECLAIM_BAD:
1086                         case -NFS4ERR_RECLAIM_CONFLICT:
1087                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1088                                 break;
1089                         case -NFS4ERR_EXPIRED:
1090                         case -NFS4ERR_NO_GRACE:
1091                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1092                         case -NFS4ERR_STALE_CLIENTID:
1093                         case -NFS4ERR_BADSESSION:
1094                         case -NFS4ERR_BADSLOT:
1095                         case -NFS4ERR_BAD_HIGH_SLOT:
1096                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1097                                 goto out_err;
1098                 }
1099                 nfs4_put_open_state(state);
1100                 goto restart;
1101         }
1102         spin_unlock(&sp->so_lock);
1103         return 0;
1104 out_err:
1105         nfs4_put_open_state(state);
1106         return status;
1107 }
1108
1109 static void nfs4_clear_open_state(struct nfs4_state *state)
1110 {
1111         struct nfs4_lock_state *lock;
1112
1113         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1114         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1115         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1116         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1117         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1118                 lock->ls_seqid.flags = 0;
1119                 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1120         }
1121 }
1122
1123 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1124 {
1125         struct nfs4_state_owner *sp;
1126         struct rb_node *pos;
1127         struct nfs4_state *state;
1128
1129         /* Reset all sequence ids to zero */
1130         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1131                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1132                 sp->so_seqid.flags = 0;
1133                 spin_lock(&sp->so_lock);
1134                 list_for_each_entry(state, &sp->so_states, open_states) {
1135                         if (mark_reclaim(clp, state))
1136                                 nfs4_clear_open_state(state);
1137                 }
1138                 spin_unlock(&sp->so_lock);
1139         }
1140 }
1141
1142 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1143 {
1144         /* Mark all delegations for reclaim */
1145         nfs_delegation_mark_reclaim(clp);
1146         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1147 }
1148
1149 static void nfs4_reclaim_complete(struct nfs_client *clp,
1150                                  const struct nfs4_state_recovery_ops *ops)
1151 {
1152         /* Notify the server we're done reclaiming our state */
1153         if (ops->reclaim_complete)
1154                 (void)ops->reclaim_complete(clp);
1155 }
1156
1157 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1158 {
1159         struct nfs4_state_owner *sp;
1160         struct rb_node *pos;
1161         struct nfs4_state *state;
1162
1163         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1164                 return 0;
1165
1166         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1167                 sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1168                 spin_lock(&sp->so_lock);
1169                 list_for_each_entry(state, &sp->so_states, open_states) {
1170                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags))
1171                                 continue;
1172                         nfs4_state_mark_reclaim_nograce(clp, state);
1173                 }
1174                 spin_unlock(&sp->so_lock);
1175         }
1176
1177         nfs_delegation_reap_unclaimed(clp);
1178         return 1;
1179 }
1180
1181 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1182 {
1183         if (!nfs4_state_clear_reclaim_reboot(clp))
1184                 return;
1185         nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1186 }
1187
1188 static void nfs_delegation_clear_all(struct nfs_client *clp)
1189 {
1190         nfs_delegation_mark_reclaim(clp);
1191         nfs_delegation_reap_unclaimed(clp);
1192 }
1193
1194 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1195 {
1196         nfs_delegation_clear_all(clp);
1197         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1198 }
1199
1200 static void nfs4_warn_keyexpired(const char *s)
1201 {
1202         printk_ratelimited(KERN_WARNING "Error: state manager"
1203                         " encountered RPCSEC_GSS session"
1204                         " expired against NFSv4 server %s.\n",
1205                         s);
1206 }
1207
1208 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1209 {
1210         switch (error) {
1211                 case -NFS4ERR_CB_PATH_DOWN:
1212                         nfs_handle_cb_pathdown(clp);
1213                         return 0;
1214                 case -NFS4ERR_NO_GRACE:
1215                         nfs4_state_end_reclaim_reboot(clp);
1216                         return 0;
1217                 case -NFS4ERR_STALE_CLIENTID:
1218                 case -NFS4ERR_LEASE_MOVED:
1219                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1220                         nfs4_state_clear_reclaim_reboot(clp);
1221                         nfs4_state_start_reclaim_reboot(clp);
1222                         break;
1223                 case -NFS4ERR_EXPIRED:
1224                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1225                         nfs4_state_start_reclaim_nograce(clp);
1226                         break;
1227                 case -NFS4ERR_BADSESSION:
1228                 case -NFS4ERR_BADSLOT:
1229                 case -NFS4ERR_BAD_HIGH_SLOT:
1230                 case -NFS4ERR_DEADSESSION:
1231                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1232                 case -NFS4ERR_SEQ_FALSE_RETRY:
1233                 case -NFS4ERR_SEQ_MISORDERED:
1234                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1235                         /* Zero session reset errors */
1236                         return 0;
1237                 case -EKEYEXPIRED:
1238                         /* Nothing we can do */
1239                         nfs4_warn_keyexpired(clp->cl_hostname);
1240                         return 0;
1241         }
1242         return error;
1243 }
1244
1245 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1246 {
1247         struct rb_node *pos;
1248         int status = 0;
1249
1250 restart:
1251         spin_lock(&clp->cl_lock);
1252         for (pos = rb_first(&clp->cl_state_owners); pos != NULL; pos = rb_next(pos)) {
1253                 struct nfs4_state_owner *sp = rb_entry(pos, struct nfs4_state_owner, so_client_node);
1254                 if (!test_and_clear_bit(ops->owner_flag_bit, &sp->so_flags))
1255                         continue;
1256                 atomic_inc(&sp->so_count);
1257                 spin_unlock(&clp->cl_lock);
1258                 status = nfs4_reclaim_open_state(sp, ops);
1259                 if (status < 0) {
1260                         set_bit(ops->owner_flag_bit, &sp->so_flags);
1261                         nfs4_put_state_owner(sp);
1262                         return nfs4_recovery_handle_error(clp, status);
1263                 }
1264                 nfs4_put_state_owner(sp);
1265                 goto restart;
1266         }
1267         spin_unlock(&clp->cl_lock);
1268         return status;
1269 }
1270
1271 static int nfs4_check_lease(struct nfs_client *clp)
1272 {
1273         struct rpc_cred *cred;
1274         const struct nfs4_state_maintenance_ops *ops =
1275                 clp->cl_mvops->state_renewal_ops;
1276         int status = -NFS4ERR_EXPIRED;
1277
1278         /* Is the client already known to have an expired lease? */
1279         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1280                 return 0;
1281         spin_lock(&clp->cl_lock);
1282         cred = ops->get_state_renewal_cred_locked(clp);
1283         spin_unlock(&clp->cl_lock);
1284         if (cred == NULL) {
1285                 cred = nfs4_get_setclientid_cred(clp);
1286                 if (cred == NULL)
1287                         goto out;
1288         }
1289         status = ops->renew_lease(clp, cred);
1290         put_rpccred(cred);
1291 out:
1292         return nfs4_recovery_handle_error(clp, status);
1293 }
1294
1295 static int nfs4_reclaim_lease(struct nfs_client *clp)
1296 {
1297         struct rpc_cred *cred;
1298         const struct nfs4_state_recovery_ops *ops =
1299                 clp->cl_mvops->reboot_recovery_ops;
1300         int status = -ENOENT;
1301
1302         cred = ops->get_clid_cred(clp);
1303         if (cred != NULL) {
1304                 status = ops->establish_clid(clp, cred);
1305                 put_rpccred(cred);
1306                 /* Handle case where the user hasn't set up machine creds */
1307                 if (status == -EACCES && cred == clp->cl_machine_cred) {
1308                         nfs4_clear_machine_cred(clp);
1309                         status = -EAGAIN;
1310                 }
1311                 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1312                         status = -EPROTONOSUPPORT;
1313         }
1314         return status;
1315 }
1316
1317 #ifdef CONFIG_NFS_V4_1
1318 void nfs41_handle_recall_slot(struct nfs_client *clp)
1319 {
1320         set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1321         nfs4_schedule_state_recovery(clp);
1322 }
1323
1324 static void nfs4_reset_all_state(struct nfs_client *clp)
1325 {
1326         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1327                 clp->cl_boot_time = CURRENT_TIME;
1328                 nfs4_state_start_reclaim_nograce(clp);
1329                 nfs4_schedule_state_recovery(clp);
1330         }
1331 }
1332
1333 static void nfs41_handle_server_reboot(struct nfs_client *clp)
1334 {
1335         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1336                 nfs4_state_start_reclaim_reboot(clp);
1337                 nfs4_schedule_state_recovery(clp);
1338         }
1339 }
1340
1341 static void nfs41_handle_state_revoked(struct nfs_client *clp)
1342 {
1343         /* Temporary */
1344         nfs4_reset_all_state(clp);
1345 }
1346
1347 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1348 {
1349         /* This will need to handle layouts too */
1350         nfs_expire_all_delegations(clp);
1351 }
1352
1353 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1354 {
1355         nfs_expire_all_delegations(clp);
1356         if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1357                 nfs4_schedule_state_recovery(clp);
1358 }
1359
1360 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1361 {
1362         if (!flags)
1363                 return;
1364         else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1365                 nfs41_handle_server_reboot(clp);
1366         else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1367                             SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1368                             SEQ4_STATUS_ADMIN_STATE_REVOKED |
1369                             SEQ4_STATUS_LEASE_MOVED))
1370                 nfs41_handle_state_revoked(clp);
1371         else if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1372                 nfs41_handle_recallable_state_revoked(clp);
1373         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1374                             SEQ4_STATUS_BACKCHANNEL_FAULT |
1375                             SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1376                 nfs41_handle_cb_path_down(clp);
1377 }
1378
1379 static int nfs4_reset_session(struct nfs_client *clp)
1380 {
1381         int status;
1382
1383         nfs4_begin_drain_session(clp);
1384         status = nfs4_proc_destroy_session(clp->cl_session);
1385         if (status && status != -NFS4ERR_BADSESSION &&
1386             status != -NFS4ERR_DEADSESSION) {
1387                 status = nfs4_recovery_handle_error(clp, status);
1388                 goto out;
1389         }
1390
1391         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1392         status = nfs4_proc_create_session(clp);
1393         if (status) {
1394                 status = nfs4_recovery_handle_error(clp, status);
1395                 goto out;
1396         }
1397         /* create_session negotiated new slot table */
1398         clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1399
1400          /* Let the state manager reestablish state */
1401         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1402                 nfs41_setup_state_renewal(clp);
1403 out:
1404         return status;
1405 }
1406
1407 static int nfs4_recall_slot(struct nfs_client *clp)
1408 {
1409         struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1410         struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1411         struct nfs4_slot *new, *old;
1412         int i;
1413
1414         nfs4_begin_drain_session(clp);
1415         new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1416                       GFP_NOFS);
1417         if (!new)
1418                 return -ENOMEM;
1419
1420         spin_lock(&fc_tbl->slot_tbl_lock);
1421         for (i = 0; i < fc_tbl->target_max_slots; i++)
1422                 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1423         old = fc_tbl->slots;
1424         fc_tbl->slots = new;
1425         fc_tbl->max_slots = fc_tbl->target_max_slots;
1426         fc_tbl->target_max_slots = 0;
1427         fc_attrs->max_reqs = fc_tbl->max_slots;
1428         spin_unlock(&fc_tbl->slot_tbl_lock);
1429
1430         kfree(old);
1431         nfs4_end_drain_session(clp);
1432         return 0;
1433 }
1434
1435 #else /* CONFIG_NFS_V4_1 */
1436 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1437 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1438 static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1439 #endif /* CONFIG_NFS_V4_1 */
1440
1441 /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1442  * on EXCHANGE_ID for v4.1
1443  */
1444 static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1445 {
1446         if (nfs4_has_session(clp)) {
1447                 switch (status) {
1448                 case -NFS4ERR_DELAY:
1449                 case -NFS4ERR_CLID_INUSE:
1450                 case -EAGAIN:
1451                         break;
1452
1453                 case -EKEYEXPIRED:
1454                         nfs4_warn_keyexpired(clp->cl_hostname);
1455                 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1456                                          * in nfs4_exchange_id */
1457                 default:
1458                         return;
1459                 }
1460         }
1461         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1462 }
1463
1464 static void nfs4_state_manager(struct nfs_client *clp)
1465 {
1466         int status = 0;
1467
1468         /* Ensure exclusive access to NFSv4 state */
1469         for(;;) {
1470                 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1471                         /* We're going to have to re-establish a clientid */
1472                         status = nfs4_reclaim_lease(clp);
1473                         if (status) {
1474                                 nfs4_set_lease_expired(clp, status);
1475                                 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1476                                                         &clp->cl_state))
1477                                         continue;
1478                                 if (clp->cl_cons_state ==
1479                                                         NFS_CS_SESSION_INITING)
1480                                         nfs_mark_client_ready(clp, status);
1481                                 goto out_error;
1482                         }
1483                         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1484                         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1485                         pnfs_destroy_all_layouts(clp);
1486                 }
1487
1488                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1489                         status = nfs4_check_lease(clp);
1490                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1491                                 continue;
1492                         if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1493                                 goto out_error;
1494                 }
1495
1496                 /* Initialize or reset the session */
1497                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1498                    && nfs4_has_session(clp)) {
1499                         status = nfs4_reset_session(clp);
1500                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1501                                 continue;
1502                         if (status < 0)
1503                                 goto out_error;
1504                 }
1505
1506                 /* First recover reboot state... */
1507                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1508                         status = nfs4_do_reclaim(clp,
1509                                 clp->cl_mvops->reboot_recovery_ops);
1510                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1511                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1512                                 continue;
1513                         nfs4_state_end_reclaim_reboot(clp);
1514                         if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1515                                 continue;
1516                         if (status < 0)
1517                                 goto out_error;
1518                 }
1519
1520                 /* Now recover expired state... */
1521                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1522                         status = nfs4_do_reclaim(clp,
1523                                 clp->cl_mvops->nograce_recovery_ops);
1524                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1525                             test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1526                             test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1527                                 continue;
1528                         if (status < 0)
1529                                 goto out_error;
1530                 }
1531
1532                 nfs4_end_drain_session(clp);
1533                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1534                         nfs_client_return_marked_delegations(clp);
1535                         continue;
1536                 }
1537                 /* Recall session slots */
1538                 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1539                    && nfs4_has_session(clp)) {
1540                         status = nfs4_recall_slot(clp);
1541                         if (status < 0)
1542                                 goto out_error;
1543                         continue;
1544                 }
1545
1546
1547                 nfs4_clear_state_manager_bit(clp);
1548                 /* Did we race with an attempt to give us more work? */
1549                 if (clp->cl_state == 0)
1550                         break;
1551                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1552                         break;
1553         }
1554         return;
1555 out_error:
1556         printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1557                         " with error %d\n", clp->cl_hostname, -status);
1558         nfs4_end_drain_session(clp);
1559         nfs4_clear_state_manager_bit(clp);
1560 }
1561
1562 static int nfs4_run_state_manager(void *ptr)
1563 {
1564         struct nfs_client *clp = ptr;
1565
1566         allow_signal(SIGKILL);
1567         nfs4_state_manager(clp);
1568         nfs_put_client(clp);
1569         module_put_and_exit(0);
1570         return 0;
1571 }
1572
1573 /*
1574  * Local variables:
1575  *  c-basic-offset: 8
1576  * End:
1577  */